Enhancing Image Clarity and Classification Through CNN-Based Dehazing with Global Memory and Attention Mechanisms

Engineering notes

Key topics: autonomous driving, perception. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

The design purpose of the proposed project involves creating an enhanced image dehazing network by introducing memory and attention regularizations into Generic Model-Agnostic Convolutional Neural Network, popularly known as GMAN, so as to enrich the visual perception of a hazy image, by appropriate classification of the images into hazy and clear. The one central aspect about this work is that a global attention mechanism in GMAN is used, which allows the model to look at interesting areas in the entire picture. This attention feature enhances the model to capture fine variance in the concentration of haze therefore improving overall classification. Moreover, using the attention module, the model will be able to emphasize important objects potentially hidden by haze resulting in more informative outputs in the form of images. The large dataset is a combination of hazy and clear images on which the model is trained to enable robustness and generalization in the real-world settings. Applied in outdoor photography, autonomous driving, video surveillance, and many computer visions tasks, the proposed project is going to present a sizeable contribution to the progress of automated and precise image quality improvement.

Links and sources

Comments